[FFmpeg-devel] [PATCH v3 7/7] avutil/la: Add function performance testing

[Rémi Denis-Courmont]

Le 24 mai 2023 10:39:59 GMT+03:00, Hao Chen <chenhao at loongson.cn> a écrit :
>
>在 2023/5/20 下午5:38, Rémi Denis-Courmont 写道:
>> Le lauantaina 20. toukokuuta 2023, 10.27.19 EEST Hao Chen a écrit :
>>> From: yuanhecai <yuanhecai at loongson.cn>
>>> 
>>> This patch supports the use of the "checkasm --bench" testing feature
>>> on loongarch platform.
>>> 
>>> Change-Id: I42790388d057c9ade0dfa38a19d9c1fd44ca0bc3
>>> ---
>>>   libavutil/loongarch/timer.h | 48 +++++++++++++++++++++++++++++++++++++
>>>   libavutil/timer.h           |  2 ++
>>>   2 files changed, 50 insertions(+)
>>>   create mode 100644 libavutil/loongarch/timer.h
>>> 
>>> diff --git a/libavutil/loongarch/timer.h b/libavutil/loongarch/timer.h
>>> new file mode 100644
>>> index 0000000000..44ed786409
>>> --- /dev/null
>>> +++ b/libavutil/loongarch/timer.h
>>> @@ -0,0 +1,48 @@
>>> +/*
>>> + * Copyright (c) 2023 Loongson Technology Corporation Limited
>>> + * Contributed by Hecai Yuan <yuanhecai at loongson.cn>
>>> + *
>>> + * This file is part of FFmpeg.
>>> + *
>>> + * FFmpeg is free software; you can redistribute it and/or
>>> + * modify it under the terms of the GNU Lesser General Public
>>> + * License as published by the Free Software Foundation; either
>>> + * version 2.1 of the License, or (at your option) any later version.
>>> + *
>>> + * FFmpeg is distributed in the hope that it will be useful,
>>> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
>>> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
>>> + * Lesser General Public License for more details.
>>> + *
>>> + * You should have received a copy of the GNU Lesser General Public
>>> + * License along with FFmpeg; if not, write to the Free Software
>>> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301
>>> USA + */
>>> +
>>> +#ifndef AVUTIL_LOONGARCH_TIMER_H
>>> +#define AVUTIL_LOONGARCH_TIMER_H
>>> +
>>> +#include <stdint.h>
>>> +#include "config.h"
>>> +
>>> +#if HAVE_INLINE_ASM
>>> +
>>> +#define AV_READ_TIME read_time
>>> +
>>> +static inline uint64_t read_time(void)
>>> +{
>>> +
>>> +#if ARCH_LOONGARCH64
>>> +    uint64_t a, id = 0;
>> Initial value is never used.
>> 
>>> +    __asm__ volatile ( "rdtime.d  %0, %1" : "=r"(a), "=r"(id) :: "memory"
>>> ); +    return a;
>>> +#else
>>> +    uint32_t a, id = 0;
>>> +    __asm__ volatile ( "rdtimel.w  %0, %1" : "=r"(a), "=r"(id) :: "memory"
>>> ); +    return (uint64_t)a;
>>> +#endif
>> Why do you clobber memory here?
>> 
>>> +}
>>> +
>>> +#endif /* HAVE_INLINE_ASM */
>>> +
>>> +#endif /* AVUTIL_LOONGARCH_TIMER_H */
>>> diff --git a/libavutil/timer.h b/libavutil/timer.h
>>> index d3db5a27ef..861ba7e9d7 100644
>>> --- a/libavutil/timer.h
>>> +++ b/libavutil/timer.h
>>> @@ -61,6 +61,8 @@
>>>   #   include "riscv/timer.h"
>>>   #elif ARCH_X86
>>>   #   include "x86/timer.h"
>>> +#elif ARCH_LOONGARCH
>>> +#   include "loongarch/timer.h"
>>>   #endif
>>> 
>>>   #if !defined(AV_READ_TIME)
>>> 
>>     Thanks for your advice.  As described in loongarch's instruction manual, the rdtime.d instruction is used as follows:
>> rdtime.d rd, rj. The rj register stores the counter ID. In this application, the value of counter ID is equal to 0.

You're setting a value, zero, to a variable `id`, that is then used as output operand. As far as the compiler is concerned, the value zero is never used and the initialisation can be elided. The value of register %1 is unspecified.

If you meant for `id` to be an input operand, the constraints are incorrect.